As part of the manufacturing process of semiconductor devices, semiconductor wafers are polished by CMP. The uniform removal of material from and the planarity of patterned and un-patterned wafers is critical to wafer process yield. Generally, the wafer to be polished is mounted on a substrate carrier which holds the wafer using a combination of vacuum suction or other means to contact the rear side of the wafer and a retaining lip or ring around the edge of the wafer to keep the wafer centered on the substrate carrier. The front side of the wafer, the side to be polished, is then contacted with an abrasive material such as an abrasive pad or abrasive strip. The abrasive pad or strip may have free abrasive fluid sprayed on it, may have abrasive particles affixed to it, or may have abrasive particles sprinkled on it.
The ideal wafer polishing process can be described by Preston's equation:
R=K.sub.p *P*V, where R is the removal rate; Kp is a function of consumables (abrasive pad roughness and elasticity, surface chemistry and abrasion effects, and contact area); P is the applied pressure between the wafer and the abrasive pad; and V is the relative velocity between the wafer and the abrasive pad. As a result, the ideal CMP process should have constant cutting velocity over the entire wafer surface, constant pressure between the abrasive pad and wafer, and constant abrasive pad roughness, elasticity, area and abrasion effects. In addition, control over the temperature and pH is critical and the direction of the relative pad/wafer velocity should be randomly distributed over the entire wafer surface.
One common type of wafer polishing apparatus is the CMP model 372M made by Westech Systems Inc. A wafer is held by a substrate carrier of the model 372M. The substrate carrier rotates about the axis of the wafer. A large circular abrasive pad is rotated while contacting the rotating wafer and substrate carrier. The rotating wafer contacts the larger rotating abrasive pad in an area away from the center of the abrasive pad.
Another related apparatus is a polishing machine for polishing semiconductor wafers containing magnetic read-write heads, disclosed in U.S. Pat. No. 5,335,453 to Baldy et al. With this machine, a semiconductor wafer is held by a substrate carrier which is moved in a circular translatory motion by an eccentric arm. The wafer is polished by contacting an abrasive strip which is advanced in one direction. The relative motion between the wafer and the abrasive strip is a combination of the circular motion of the wafer and the linear motion of the advancing abrasive strip.
While the precessing circle polishing pattern should provide more uniform velocities such that different points on the wafer see similar velocities at any given time, the velocities are still not constant. Assuming the rotation of the eccentric arm is held to a constant angular speed, the precessing circle relative motion results in fluctuating velocities. When the wafer is rotating away from the precessing direction the net relative velocity is lower, and when the wafer is rotating with precessing direction the net relative velocity is higher.
The gimbal point of a CMP substrate carrier is a critical element of the polishing process. The substrate carrier must align itself to the polish surface precisely to insure uniform, planar polishing results. Many CMP substrate carriers currently available yield wafers having anomalies in planarity. The vertical height of the pivot point above the polishing surface is also important, since the greater the height, the larger the moment that is induced about the pivot point during polishing. Two pervasive problems that exist in most CMP wafer polishing apparatuses are underpolishing of the center of the wafer, and the inability to adjust the control of wafer edge exclusion as process variables change.
For example, substrate carriers used on many available CMP machines experience a phenomenon known in the art as "nose diving". During polishing, the head reacts to the polishing forces in a manner that creates a sizable moment, which is directly influenced by the height of the gimbal point, mentioned above. This moment causes a pressure differential along the direction of motion of the head. The result of the pressure differential is the formation of a standing wave of the chemical slurry that interfaces the wafer and the abrasive surface. This causes the edge of the wafer which is at the leading edge of the substrate carrier, to become polished faster and to a greater degree than the center of the wafer.
The removal of material on the wafer is related to the chemical action of the slurry. As slurry is inducted between the wafer and the abrasive pad and reacts, the chemicals responsible for removal of the wafer material gradually become exhausted. Thus, the removal of wafer material further from the leading edge of the substrate carrier (i.e., the center of the wafer) experiences a diminished rate of chemical removal when compared with the chemical action at the leading edge of the substrate carrier (i.e., the edge of the wafer), due to the diminished activity of the chemicals in the slurry when it reaches the center of the wafer. This phenomenon is sometimes referred to as "slurry starvation".
Apart from attempts to reshape the crown of the substrate carrier, other attempts have been made to improve the aforementioned problem concerning "nose diving". In a prior art substrate carrier that gimbals through a single bearing at the top of the substrate carrier, sizable moments are generated because the effective gimbal point of the substrate carrier exists at a significant, non-zero distance from the surface of the polishing pad. Thus, the frictional forces, acting at the surface of the polishing pad, act through this distance to create the undesirable moments.
U.S. Pat. No. 5,377,451 to Leoni et al. describes a wafer carrier that "projects" the effective gimbal point down to the surface of the polishing pad, thereby eliminating the moment arm through which the frictional forces create the undesirable "nose diving". Leoni et al. produce this effect by instituting a conical bearing assembly which allows the projection of a "universal pivot point" to a point that is located at or near the surface of the polishing surface. The solution proposed by Leoni et al., however, requires the use of a number of bearings in the assembly in order to effect this projection, thereby increasing the cost of the wafer carrier. Additionally, there is still a moment produced because of the actual contact points at the bearings. There is also a substantial risk that, due to inexact manufacturing, the projected pivot point will not lie exactly on the contact surface of the carrier, which will also introduce moments.